Hi,
Filed a bug to track for future reference:
https://bugs.openjdk.java.net/browse/JDK-8087102
Actually, this is not a bug. Two things:
1) --compile-only / -co option is no longer mandatory to use
--debug-dump-dir option/
2) optimistic types is off by default in jdk8u and it is on by default
with jdk9.
With optimistic types, code generation is lazier & optimistic -- code is
generated/re-generated as you run the code. With jdk9, you can either
1) Turn off optimistic types with jdk9 (using -ot=false option) and dump
similar code as in jdk8u
2) or run the code (i.e., remove -co / --compile-only option) and get
all generated (and regenerated) code dumped.
PS. I've resolved the bug as "not an issue".
Hope this explains,
-Sundar
On Wednesday 10 June 2015 02:55 AM, Chris Newland wrote:
Hi,
Has Nashorn's behaviour for dumping bytecode changed in JDK9?
Given this JavaScript:
https://gist.github.com/chriswhocodes/bfe139592adb846967d8
Executed using
$JAVA_HOME/bin/jjs -co --dump-debug-dir=. run.js
Under 8u60b18 the bytecode is produced as expected:
https://gist.github.com/chriswhocodes/e3a85c3613b084e45356
This visualises correctly in JITWatch[1] and the bytecode maps back to the
source as expected.
But under 9b67 only a small amount of bytecode is output that doesn't
represent the source program:
https://gist.github.com/chriswhocodes/c6c6e33e7994c98d4975
The correct program output is produced under both executions so I don't
think the difference is down to dead code elimination.
Possible regression?
Thanks,
Chris
@chriswhocodes
[1] https://github.com/AdoptOpenJDK/jitwatch
